Surface Capture for Performance-Based Animation

这篇文章主要讲从不同view的images如何重建三维模型

首先用绿屏提取每张图片的轮廓, 然后用shape from silhouette建立visual hull

7 Finally, we reconstruct the scene as the surface within the visual hull that passes through the surface features while reproducing the silhouette images and maximizing the consistency in appearance between views.
为了进一步refine这个visual hull, 找出相邻两个视角下的feature, 让reconstructed surface pass through these features

dense reconstrction
我估计重建的时候visual hull作为source, 然后merged multiple-view stereo 做sink, 然后用最小割最大流提取最后的surface

然后后面就是remesh和texture recovery, 没怎么细看

总的来说, 这篇文章写得很笼统, 很少技术细节没有写到, 复现起来有一定难度